Optimization of a gas turbine model combustor due to variations in geometrical characteristics of stabilizing air jets

Abstract

• Interaction of input parameters (diameter, angle, and position) is considered. • ANN is used to extract fitness (trained) function for GA. • NOx, CO, and PF are improved by up to 11.31%, 2.88%, and 21.07%, respectively. • Soot is decreased by up to 24.46%. The main purpose of this paper is to optimize the objective parameters in a gas turbine model combustor. These parameters include N O x and C O pollutants as well as pattern factor. C O is based on a maximum produced value in the model combustor. To optimize the objective parameters, the input characteristics (such as diameter, angle, and position of the stabilizing air jets) are varied. Simulation of the combustion process comprises turbulent flow, combustion, radiative heat transfer, and fuel injection modeling. The models employed for this simulation, based on RANS approach, are realizable k - ε for flow turbulence, steady flamelet model for combustion, discrete ordinates model for radiative heat transfer, and Eulerian-Lagrangian approach for fuel injection. A diesel fuel (C 10 H 22 ) is used in the model combustor and N O x modeling is conducted via post-processing. Artificial neural network is applied to gain an approximated trained function based on results and input geometrical characteristics. Data for training the artificial neural network is generated by means of design of experiments. For finding the optimum point, genetic algorithm is utilized. The obtained neural network function is applied to genetic algorithm so as to extract the optimum points. Since there is more than one objective parameter a set of optimum points is obtained which is called the Pareto front. Using multiple-criteria decision making method (LINMAP model), the final optimum point is achieved. The optimum point shows that the amounts of N O x , C O , and pattern factor are improved by up to 11.31 %, 2.88 %, and 21.07 %, respectively. The value of soot which is not a part of optimization is also improved by up to 24.46 %.